Top AI models and providers
We don't vote here. This page puts four public leaderboards side by side: people's blind votes on Arena, its WebDev board for building websites, the Artificial Analysis intelligence index, and what developers actually spend tokens on, from OpenRouter. Every number links to where it came from.
#1 on WebDev Arena, 1824. source
#1 on the Arena text board, 1526 (preliminary). source
65 on the Artificial Analysis index. source
61.9T tokens a week on OpenRouter. source
Index 46, the top open model. source
Top models for building websites
Ordered by WebDev Arena, because that is the job openconstructor gives them. The other columns show the same model on the other boards; a dash means it is not in that board's top list. Each model is listed once, at its best setting.
| # | Model | WebDev Arena | Arena text | AA index | OpenRouter | Use it via |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2Anthropic | 1824 | — | 65 | — | |
| 2 | Claude Opus 5.5Anthropic | 1818 | 1507 | 59 | #7 | |
| 3 | GPT-6.2 AstraOpenAI | 1781 | — | 62 | — | |
| 4 | Claude Sonnet 5.5Anthropic | 1773 | — | 55 | — | |
| 5 | Gemini 4 CarbonGoogle | 1769 | 1526 | 60 | — | |
| 6 | GPT-6.2 SolOpenAI | 1757 | — | 57 | — | |
| 7 | Qwen4-MaxAlibaba | 1750 | — | 43 | — | |
| 8 | GPT-6.1 SolOpenAI | 1744 | — | — | — | |
| 9 | Muse WonderMeta | 1723 | 1511 | 50 | — | no public API in our list |
| 10 | Gemini 4 HydrogenGoogle | 1703 | 1490 | 54 | — | |
| 11 | Kimi K3.1 ProMoonshot | 1684 | — | 46 | — |
Sources: WebDev Arena (Nov 18) · Arena text (Nov 11, 9.1M votes) · Artificial Analysis · OpenRouter (week to Nov 17). "Use it via" lists the first-party API and the services in openconstructor that serve it.
Arena — blind votes
Two anonymous answers, people pick the better one. Elo-style score.
| # | Model | Maker | Score |
|---|---|---|---|
| 1 | Gemini 4 Carbon preliminary | 1526 | |
| 2 | Gemini 4 Argon preliminary | 1522 | |
| 3 | Muse Wonder | Meta | 1511 |
| 4 | Claude Opus 5.5 | Anthropic | 1507 |
| 5 | Claude Fable 5 | Anthropic | 1504 |
| 6 | Claude Opus 4.6 | Anthropic | 1503 |
| 7 | Claude Fable 5.1 | Anthropic | 1502 |
| 8 | Claude Opus 4.7 | Anthropic | 1501 |
| 9 | Gemini 3.8 Flash preliminary | 1497 | |
| 10 | Muse Spark 1.3 | Meta | 1497 |
arena.ai/leaderboard/text · updated Nov 11, 2026
Artificial Analysis — intelligence
A fixed set of hard evals. Cost and time are per full run.
| # | Model | Index | Cost | Time |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 Max | 65 | $26.54 | 41.1 min |
| 2 | GPT-6.2 Astra Max | 62 | $4.00 | 12.9 min |
| 3 | Gemini 4 Carbon High | 60 | $7.76 | 34.5 min |
| 4 | Claude Opus 5.5 Max | 59 | $10.04 | 24.6 min |
| 5 | GPT-6.2 Sol Max | 57 | $1.01 | 8.0 min |
| 6 | Claude Sonnet 5.5 Max | 55 | $10.02 | 32.2 min |
| 7 | Gemini 4 Hydrogen Xhigh | 54 | $3.16 | 15.0 min |
| 8 | GPT-6.1 Terra Max | 51 | $0.96 | 14.5 min |
| 9 | Gemini 4 Argon High | 50 | $4.97 | 29.3 min |
| 10 | Kimi K3.1 Pro | 46 | $3.68 | 38.4 min |
artificialanalysis.ai · copied Nov 18, 2026
What people actually use
Tokens processed on OpenRouter in a week. Cheap, fast models win here — real apps send a lot of small requests.
| # | Model | Maker | Tokens | Change |
|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017 | DeepSeek | 61.9T | +15% |
| 2 | Hy4 Flash | Tencent | 22.8T | new |
| 3 | DeepSeek V4.1 Flash | DeepSeek | 15.9T | −6% |
| 4 | Hy4 | Tencent | 14.3T | −24% |
| 5 | GLM 5.3 Flash | Z.ai | 11.5T | +2% |
| 6 | GPT-6 Luna | OpenAI | 8.38T | +19% |
| 7 | Claude Opus 5.5 | Anthropic | 7.53T | +11% |
| 8 | Step 5 Preview | StepFun | 6.57T | −5% |
| 9 | MiMo-V2.6-Flash | Xiaomi | 4.11T | +44% |
| 10 | GLM 5.3 | Z.ai | 3.62T | −1% |
openrouter.ai/rankings · 7 days to Nov 17, 2026
Top providers
Free, without any key
Work in openconstructor straight away, no sign-up. Slower and smaller models than the top of the boards — our tests show which ones build decent pages.
Free with a key
A free account and an API key; daily or per-minute limits.
Cheapest and fastest hosts for open models
Open models run on many clouds at different prices. Artificial Analysis runs the same eval on each host; cost and time are for that full run.
| Model | Cheapest | Fastest | Good balance |
|---|---|---|---|
| Kimi K3.1 Pro | PrimaLabs — $2.94 / 9.9 min | Lithos AI (Ultra) — 4.3 min / $6.32 | Baseten — $3.26 / 7.1 min |
| DeepSeek V4.1 Flash 1017 | DeepInfra — $0.19 / 18.0 min | Baseten (Fast) — 2.6 min / $2.32 | GMI Cloud — $0.30 / 10.8 min |
| GLM-5.3 | Multiverse Computing — $1.48 / 18.8 min | Supernova (FP4, Fast) — 5.1 min / $13.49 | Bitdeer AI — 6.9 min / $3.59 |
artificialanalysis.ai/leaderboards/providers · copied Nov 18, 2026
How this page is made
Numbers are copied by hand from the four public pages on the dates shown and are not adjusted. Boards change every few days, so check the source before a big decision. "Preliminary" means Arena has too few votes yet for a stable score. Model names are the makers' own; settings such as "High" or "Max" are folded into one row per model. Something wrong or out of date? Tell us on the changelog page.